Pyspark 将字符串列表转换为 ArrayType()

Question

我是 pyspark 的新手，我可以使用一些指导。 所以我正在处理一些文本数据，最终我想摆脱在整个语料库中出现频率不够或出现频率过高的单词。

数据看起来像这样，每一行都包含一个句子：

+--------------------+
|             cleaned|
+--------------------+
|China halfway com...|
|MCI overhaul netw...|
|script kiddy join...|
|look Microsoft Mo...|
|Americans appear ...|
|Oil Eases Venezue...|
|Americans lose be...|
|explosion Echo Na...|
|Bush tackle refor...|
|jail olympic pool...|
|coyote sign RW Jo...|
|home pc key Windo...|
|bomb defuse Blair...|
|Livermore   need ...|
|hat ring fast Wi ...|
|Americans dutch s...|
|Insect Vibrations...|
|Britain sleepwalk...|
|Ron Regan Jr Kind...|
|IBM buy danish fi...|
+--------------------+

所以基本上我使用split()从pyspark.sql.functions拆分字符串，然后计算每个单词的出现次数，提出一些标准并创建需要删除的单词列表。

然后我使用以下功能

from pyspark.sql.functions import udf
from pyspark.sql.types import *


def remove_stop_words(list_of_tokens, list_of_stopwords):
    '''
    A very simple fuction that takes in a list of word tokens and then gets rid of words that are in stopwords list
    '''
    return [token for token in list_of_tokens if token not in list_of_stopwords]

def udf_remove_stop_words(list_of_stopwords):
    '''
    creates a udf that takes in a list of stop words and passes them onto remove_stop_words
    '''
    return udf(lambda x: remove_stop_words(x, list_of_stopwords))
 
wordsNoStopDF = splitworddf.withColumn('removed', udf_remove_stop_words(list_of_words_to_get_rid)(col('split')))

其中list_of_words_to_get_rid是我试图摆脱的单词列表，该管道的输入如下所示

+--------------------+
|               split|
+--------------------+
|[China, halfway, ...|
|[MCI, overhaul, n...|
|[script, kiddy, j...|
|[look, Microsoft,...|
|[Americans, appea...|
|[Oil, Eases, Vene...|
|[Americans, lose,...|
|[explosion, Echo,...|
|[Bush, tackle, re...|
|[jail, olympic, p...|
+--------------------+
only showing top 10 rows

并且输出看起来像以下带有相应架构的

+--------------------+--------------------+
|               split|             removed|
+--------------------+--------------------+
|[China, halfway, ...|[China, halfway, ...|
|[MCI, overhaul, n...|[MCI, overhaul, n...|
|[script, kiddy, j...|[script, join, fo...|
|[look, Microsoft,...|[look, Microsoft,...|
|[Americans, appea...|[Americans, appea...|
|[Oil, Eases, Vene...|[Oil, Eases, Vene...|
|[Americans, lose,...|[Americans, lose,...|
|[explosion, Echo,...|[explosion, Echo,...|
|[Bush, tackle, re...|[Bush, tackle, re...|
|[jail, olympic, p...|[jail, olympic, p...|
|[coyote, sign, RW...|[coyote, sign, Jo...|
|[home, pc, key, W...|[home, pc, key, W...|
|[bomb, defuse, Bl...|[bomb, defuse, Bl...|
|[Livermore, , , n...|[Livermore, , , n...|
|[hat, ring, fast,...|[hat, ring, fast,...|
|[Americans, dutch...|[Americans, dutch...|
|[Insect, Vibratio...|[tell, Good, Time...|
|[Britain, sleepwa...|[Britain, big, br...|
|[Ron, Regan, Jr, ...|[Ron, Jr, Guy, , ...|
|[IBM, buy, danish...|[IBM, buy, danish...|
+--------------------+--------------------+

root
 |-- split: array (nullable = true)
 |    |-- element: string (containsNull = true)
 |-- removed: string (nullable = true)

所以我的问题是如何将removed的列变成像split这样的数组？ 我希望使用explode来计算单词出现次数，但我似乎无法弄清楚该怎么做。 我尝试使用regex_replace去掉括号，然后用,作为要拆分的模式拆分字符串，但这似乎只向列remove添加了一个括号remove 。

我是否可以对我使用的函数进行一些更改，让它们返回一个字符串数组，如列split 。

这里的任何指导将不胜感激！

Answer 1

您还没有为您的 UDF 定义返回类型， 默认情况下为StringType ，这就是您removed列是字符串的原因。 您可以像这样添加使用返回类型

from pyspark.sql import types as T

udf(lambda x: remove_stop_words(x, list_of_stopwords), T.ArrayType(T.StringType()))

Answer 2

您可以更改 UDF 的返回类型。 但是，我建议不要使用任何splited从类型数组splited的列中删除单词list_of_words_to_get_rid列表，因为您可以简单地使用 spark 内置函数array_except 。

下面是一个例子：

import pyspark.sql.functions as F

df = spark.createDataFrame([("a simple sentence containing some words",)], ["cleaned"])

list_of_words_to_get_rid = ["some", "a"]

wordsNoStopDF = df.withColumn(
    "split",
    F.split("cleaned", " ")
).withColumn(
    "removed",
    F.array_except(
        F.col("split"),
        F.array(*[F.lit(w) for w in list_of_words_to_get_rid])
    )
).drop("cleaned")

wordsNoStopDF.show(truncate=False)
#+----------------------------------------------+-------------------------------------+
#|split                                         |removed                              |
#+----------------------------------------------+-------------------------------------+
#|[a, simple, sentence, containing, some, words]|[simple, sentence, containing, words]|
#+----------------------------------------------+-------------------------------------+